Recently, training deep neural networks (DNNs) on edge devices has attracted much attention due to its strong adaptability and avoidance of private data transmission. However, limited computational, storage, and energy resources pose significant challenges for edge devices. The structural and computational redundancies in DNNs create opportunities for sparse training through model pruning and zero-computation skipping. Although feasible, the sparse training accelerator design encounters common issues, such as redundant data duplication and unbalanced workloads, caused by irregular sparsity. To address these issues, this paper proposes a sparse DNN training accelerator, SDTA, together with a hierarchical pre-fetching buffer and a dynamic scheduler to achieve high design efficiency. The SDTA is deployed on the FPGA XCVU3P platform. Compared to the prior FPGA-based accelerators and the GPU, SDTA improves the energy efficiency by up to 2.29×, the storage utilization efficiency by up to 7.37×, and the computational efficiency by up to 1.9×. Compared to the dense accelerator, it achieves a speedup of up to 5.88×, while ensuring model accuracy.

This paper introduces a fault diagnosis method for synchronous traction machine demagnetization based on the Informer model. The method utilizes the Informer model to analyze and model the sensor data of the synchronous traction machine, adaptively learn the feature representation of time series data, and predict the future states in order to accurately identify and predict demagnetization faults. The specific steps include: (1) collecting sensor data of the synchronous traction machine, including parameters such as current, voltage, and rotational speed; (2) preprocessing the data, including denoising, normalization, and feature extraction; (3) constructing the Informer model and training and optimizing it using the preprocessed sensor data; and (4) using the trained model to predict and determine new sensor data, thereby achieving an accurate diagnosis of demagnetization faults. The advantages of this method are as follows: (1) automatic learning and extraction of important features of time series data without the need for manual feature design, improving diagnostic accuracy; (2) ability to handle long sequence data and strong modeling capability for time dependencies, better predicting future states and fault occurrences; and (3) adaptability to the characteristics and data features of different elevator systems and strong generalization capability.

In this paper, an interpolation-free fast factorized back-projection (IF-FFBP) algorithm is proposed for high-resolution spotlight synthetic aperture radar (SAR) processing. Different from the original FFBP utilizing two-dimensional image-domain interpolation for sub-aperture fusion, IF-FFBP finishes the image merging steps using chirp-z transform and circular shifting. Under the restriction of the applicable scope, IF-FFBP yields enhanced efficiency over the 4 times upsampling interpolation based FFBP, and keeps the high precision simultaneously. Finally, Real-data experiment verifies the efficiency superiorities of the FIM-FFBP.